Table 4 Event-based sequence classification on irregularly sequential MNIST

From: Closed-form continuous-time neural networks

Model

Accuracy (%)

Time per epoch (min)

ODE-RNN7

72.41 ± 1.69

14.57

CT-RNN48

72.05 ± 0.71

17.30

Augmented LSTM44

82.10 ± 4.36

2.48

CT-GRU49

87.51 ± 1.57

3.81

RNN-Decay7

88.93 ± 4.06

3.64

Bi-directional RNN7

94.43 ± 0.23

8.097

GRU-D51

95.44 ± 0.34

3.42

PhasedLSTM52

86.79 ± 1.57

5.69

GRU-ODE7

80.95 ± 1.52

6.76

CT-LSTM50

94.84 ± 0.17

3.84

coRNN57

94.44 ± 0.24

3.90

Lipschitz RNN58

95.92 ± 0.16

3.86

ODE-LSTM9

95.73 ± 0.24

6.35

Cf-S (current work)

95.23 ± 0.16

2.73

CfC-noGate (current work)

96.99 ± 0.30

3.36

CfC (current work)

95.42 ± 0.21

3.62

CfC-mmRNN (current work)

98.09 ± 0.18

5.50

  1. Test accuracy shown as mean ± s.d. (n = 5). Bold values indicate the highest accuracy and best time per epoch (min).
Back to article page